Table of contents
Three technical schools of privacy computing: federated learning, secure multi-party computing, and trusted computing
1. Multi-party secure computing, first proposed by Academician Yao Qizhi in 1982, through the design of special encryption algorithms and protocols, the problem of secure computing agreed functions in the absence of a trusted third party. In recent years, based on homomorphic encryption, secret sharing, Basic technologies such as oblivious transmission and obfuscated circuits, and multi-party secure computing of protocols are gradually being applied (in classic multi-party secure computing, two-party computing mainly adopts the scheme of oblivious transmission and obfuscated circuits, and more than three parties further combine secret sharing , so there are also views that Homomorphic encryption as a privacy computing technology independent of multi-party secure computing ).
2. Federated learning is a distributed machine learning framework that can realize computational training tasks such as joint modeling without sharing the original data of all parties, and ensure data security and control while breaking data islands. According to different types of computing data sets, it can be divided into horizontal and vertical federated learning and federated migration learning .
3. Trusted execution environment refers to a secure isolation environment that runs on trusted hardware , executes authorized security software, and ensures the confidentiality and integrity of key codes and data from being damaged by malicious software . Although strictly speaking, the trusted execution environment does not realize "data is available and invisible", but because of its high versatility, low development difficulty, and more flexibility to adapt to various complex algorithms than multi-party secure computing, it is also regarded as Computing means for data privacy.
Different technical paths have different security levels and applicable scenarios.
The first two calculations are based on pure software and cryptographic algorithms. Multi-party secure computing is subject to the complexity of cryptographic algorithm design and calculation . The current performance is relatively low and development is difficult. However, the research enthusiasm in the academic field is high, and the future development speed will be relatively fast. quick.
Federated learning has good performance, but currently constrained by the framework of machine learning algorithms, it is difficult to design methods for problems that need to be solved in specific scenarios. It is generally used to support relatively simple operation logic.
Trusted computing relies on a secure and trusted isolation environment built by third-party hardware manufacturers, and has good performance and algorithm adaptability. However, it needs to be combined with the first two technologies to achieve privacy computing that does not rely on trusted third parties in the true sense.
3. The integration and development of privacy computing technology is a trend, and it is also an examination paper to test the skill of manufacturers .
Among the three technical schools of privacy computing, multi-party secure computing focuses on security, federated learning focuses on efficiency, and trusted execution environment focuses on generalization. With the development and application of privacy computing technology, as well as the continuous exploration of research institutions and manufacturers in the field, privacy computing has The genre of similar technologies is gradually moving towards integration, and the application of some new technologies such as blockchain has also been added to the field of privacy computing.
At present, manufacturers focusing on the field of privacy computing generally combine two or three of the three technology paths to achieve complementary advantages and flexibly adapt to application scenarios in different industries. It can be said that only when the in-depth integration of privacy computing technology is truly realized and the innovative combination with new technologies in related fields can it be called a senior player in this field.
Bafen, as a comprehensive player in the field of privacy computing, has embarked on a path that few people have taken.
Baquan's privacy computing system is independently developed based on core technologies such as federated learning, secure multi-party computing, trusted execution environment, and blockchain, and can provide privacy protection capabilities for data exchange and data sharing. Below, let’s take a closer look at how 8Wen integrates these technologies to launch 8Wen’s privacy computing system and cross-chain big data platform.
Each computing participant needs to deploy a privacy computing system in the security area of its own data center, and establish a multi-party distributed privacy computing mechanism. Each participant publishes data description information and computing logs to the blockchain for traceability and evidence collection.
In an eight-component privacy computing system,
Multi-party secure computing is mainly to realize the joint computing of multi-party data under the premise of ensuring data security ;
Federated learning is mainly to complete joint machine learning tasks on the premise that the original private data of multiple participants does not leave the private security boundary of each party ;
The trusted execution environment is used to ensure that authorized programs are run in a secure area built on trusted hardware ;
In addition, the blockchain is used to record the calculation process and results, realize the whole process of credible traceability and data rights certification, and ensure the rights and interests of all participants under the condition of "data invisible".
Through the above process, Eight Elements has independently developed a high-performance, high-adaptation, high-security, and high-credibility privacy computing solution . At present, customers in various fields, including finance, have expressed interest in this privacy computing solution.